Kravexo

hyperfixation on AI | always dyor

[Kravexo @kravexo]

@0x_Renjii you are right

[Kravexo @kravexo]

Someone turned Meta Ray-Bans into a 3,000-Elo chess cheat.

And the AI isn’t the one playing.

Here’s the setup:

→ The glasses video-call his own Mac on WhatsApp. The Mac sees exactly what he sees.
→ Every 1–3 seconds, the Mac grabs a frame straight from the call window.
→ A cheap check runs first: is the board visible? A hand in the way? Did a piece move? Nothing changed → the expensive model stays asleep.
→ Claude Opus 5.5 reads the board, compares it to the last position, and only accepts a legal move.
→ Stockfish finds the reply in half a second, and macOS speaks it into the call: “Rook to d1.”
The 3,000 Elo comes from Stockfish, a classic chess engine. Opus 5.5 only reads the board. The cheap filter in front of it saves money on every frame.
Same playbook works in business:
Cheap filter first.
Expensive model only when it matters.
Hard math goes to regular software.
In chess, of course, it’s cheating. Tournaments will have to check players’ glasses now

[ClaudeDevs @ClaudeDevs]

Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users.

Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.

[Kravexo @kravexo]

$100 to $75,796 in 11 days is the headline.

The line I’d actually keep from this guide: “If it says NO EDGE, believe it.”

Every strategy gets tested on years it never saw, and anything that can’t beat the benchmark is deleted the same day.

[savip. @savipww]

https://t.co/YYQoOfAkKF

[Kravexo @kravexo]

"$50,000 a month on the table" ≠ $50,000 a month.

Quick math, generous assumptions:
25 clinics
→ 20% reply = 5 conversations
→ 25% close = 1 client
→ $2,000/mo

Still great for one night of an agent's work.
But that's the gap between a screenshot and a business.

Same in AI trading: the "$3K/day bot" is the table.
Fees, slippage and drawdowns are the math.

What reply rate do you actually get on cold outreach?

[Anatoli Kopadze @AnatoliKopadze]

Sam Altman after finding out a lab nobody was watching just built a model 100x faster and 150x cheaper than GPT-6 Astra at engineering design.

[Kravexo @kravexo]

Anthropic quietly shipped an advisor mode for Claude Code and it's ACTUAL F*CKING GOLD

Run Sonnet, Haiku and Opus 5.5 at the same time, each locked into the lane it dominates

Opus-grade intelligence on critical architectural calls, zero Opus burn on routine terminal iterations

Launch Claude Code with this flag pair, then add Haiku subagents with the prompt below:

claude --model sonnet --advisor opus

how the workload divides:

- Sonnet 5.5 leads the main session and cuts the actual code
- Haiku 5.5 swarms repos, greps call sites, and parses docs concurrently
- Opus 5.5 stays on standby as advisor for high-stakes checkpoints
- you stop burning tokens on work that never needed frontier reasoning

Opus mostly gets called at three points:

- right before locking an architectural plan
- the second the same error hits twice
- right before marking a complex task complete

Sonnet builds, Haiku explores, Opus reviews

JEV engineering applies the exact same logic one layer deeper: deterministic branches (file selection, tool dispatch, fast retry vs abort) clear in sub-milliseconds, so the frontier models only handle genuine forks

the architecture tree:
- Sonnet 5.5 drives the primary terminal session and edits code
- search crawls repo structure on Haiku 5.5
- grep locates call signatures on Haiku 5.5
- docs fetches API specs and schemas on Haiku 5.5
- Opus 5.5 stands on call as advisor, ingesting full context when pinged
- JEV resolves instant deterministic routing

feed this tree and prompt directly into Claude Code:

"Configure my Claude Code environment around this multi-model tree:
1. Run the primary session on Sonnet 5.5 with Opus 5.5 as advisor

2. Spawn three read-only Haiku 5.5 subagents: search (repo mapping), grep (call sites), docs (API references)

3. Enforce one rule in CLAUDE.md: ping advisor before locking any plan, upon repeating any error twice, and before final task handoff

4. Route instantaneous deterministic forks (tool choice, path check, retries) through JEV

Show every file update as a diff first, make zero changes without confirmation"

[Kravexo @kravexo]

Quick check for any strategy:
Break-even ratio = (1 − win rate) ÷ win rate

50% → wins must be ≥ 1× losses
40% → ≥ 1.5×
30% → ≥ 2.3×

Then subtract fees. Most "profitable" bots die right there.

[Kravexo @kravexo]

A trading bot that loses 62% of its trades can make money.
One that wins 90% can go broke.

The math:
38% win rate → profitable if wins are 1.6× your losses
90% win rate → one loss 9× your avg win wipes you out

AI doesn't change this. It just runs it faster.

Next time you see a "$3K/day bot" screenshot, ask one thing: avg win vs avg loss, after fees.

What's the first number you check?